ArchiGAN: Artificial Intelligence x
Architecture
Stanislas Chaillou
AI will soon massively empower architects in their day-to-day practice. This article
provides a proof of concept. The framework used here offers a springboard for
discussion, inviting architects to start engaging with AI, and data scientists to consider
Architecture as a field of investigation. In this article, we summarize a part of our
thesis, submitted at Harvard in May 2019, where Generative Adversarial Neural
Networks (or GANs) get leveraged to design floor plans and entire buildings (Fig. 1).
We believe that a statistical approach to design conception will shape AI’s potential
for Architecture. This approach is less deterministic and more holistic in character.
Rather than using machines to optimize a set of variables, relying on them to extract
significant qualities and mimicking them all along the design process represents a
paradigm shift.
We can unpack floor plan design into three distinct steps:
(I) building footprint massing
(II) program repartition
(III) furniture layout
Each step corresponds to a Pix2Pix GAN-model trained to perform one of the
three tasks above. By nesting these models one after the other, we create an entire
apartment building “generation stack” while allowing for user input at each step.
Additionally, by tackling multi-apartment processing, this project scales beyond the
simplicity of single-family houses.
Beyond the mere development of a generation pipeline, this attempt aims at
demonstrating the potential of GANs for any design process, whereby nesting
GAN models, and allowing user input between them, we try to achieve a back and
forth between humans and machines, between disciplinarian intuition and technical
innovation.
S. Chaillou (B)
Spacemaker, Oslo, Norway
e-mail: stan@spacemaker.ai
© Springer Nature Singapore Pte Ltd. 2020
P. F. Yuan et al. (eds.), Architectural Intelligence,
https://doi.org/10.1007/978-981-15-6568-7_8
117
118
S. Chaillou
Fig. 1 GAN-Generated masterplan. Source Author
1 Representation, Learning, and Framework
Pix2Pix uses a conditional Generative Adversarial Network (cGAN) to learn a
mapping from an input image to an output image. The network consists of two main
pieces, the Generator and the Discriminator. The Generator transforms the input
image to an output image; the Discriminator tries to guess if the image was produced
by the generator or if it is the original image. The two parts of the network challenge
each other resulting in higher quality outputs which are difficult to differentiate from
the original images.
We use this ability to learn image mappings which lets our models learn topological features and space organization directly from floor plan images. We control the
type of information that the model learns by formatting images. As an example, just
showing our model the shape of a parcel and its associated building footprint yields
a model able to create typical building footprints given a parcel’s shape.
We used Christopher Hesse’s implementation of Pix2Pix. Figure 2 displays the
results of a typical training. We show there how one of my GAN-models progressively
learns how to layout rooms and the position of doors and windows in space–also
called fenestration–for a given apartment unit in the sequence in Fig. 2. Although the
initial attempts proved imprecise, the machine builds some form of intuition after
250 iterations.
2 Precedents
The early work of Isola et al. in November 2018 enabling image-to-image translation
with their model Pix2Pix has paved the way for our research.
ArchiGAN: Artificial Intelligence x Architecture
119
Fig. 2 Apartment architectural sequence. Source Author
Zheng and Huang in 2018 first studied floor plan analysis using GAN. The authors
proposed to use GANs for floor plan recognition and generation using Pix2PixHD. [1]
Floor plan images processed by their GAN architecture get translated into programmatic patches of colors. Inversely, patches of colors in their work turn into drawn
rooms. If the user specifies the position of openings and rooms, the network elements
laid out become furniture.
Nathan Peters’ thesis [2] at the Harvard Graduate School of Design in the same
year tackled the possibility of laying out rooms across a single-family home footprint.
Peters’ work turns an empty footprint into programmatic patches of color without
specified fenestration.
Regarding GANs as design assistants, Nono Martinez’ thesis [3] at the Harvard
GSD in 2017 investigated the idea of a loop between the machine and the designer
to refine the very notion of “design process”.
3 Stack and Models
We build upon the previously described precedents to create a three-step generation
stack. As described in Fig. 3, each model of the stack handles a specific task of the
workflow: (I) footprint massing, (II) program repartition, (III) furniture layout.
An architect is able to modify or fine-tune the model’s output between each step,
thereby achieving the expected machine-human interaction.
120
S. Chaillou
Fig. 3 Generation stack in three models. Source Author
3.1 Model I: Footprint
Building footprints significantly define the internal organization of floor plans. Their
shape is heavily conditioned by their surroundings and, more specifically, the shape
of their parcel. Since the design of a housing building footprint can be inferred from
the shape of the piece of land it stands on, we have trained a model to generate
typical footprints, using GIS-data (Geographic Information System) from the city of
Boston. We feed pairs of images to the network during training in a format suitable
for Pix2Pix, displaying the raw parcel (left image) and the same parcel with a given
building drawn over it (right image). We show some typical results in Fig. 4.
Fig. 4 Results of Model I. Source Author
ArchiGAN: Artificial Intelligence x Architecture
121
Fig. 5 Results of model II. Source Author
3.2 Model II: Program
Model II handles repartition and fenestration. The network takes as input the footprint
of a given housing unit produced by Model I, the position of its entrance door (green
square), and the position of the main windows specified by the user (red lines). The
plans used to train the network to derive from a database of 800 + plans of apartments,
properly annotated and given in pairs to the model during training. In the output, the
program encodes rooms using colors while representing the wall structure and its
fenestration using a black patch. Some typical results are displayed in Fig. 5.
3.3 Model III: Furnishing
Finally, Model III tackles the challenge of furniture layout using the output of model
II. This model trains on pairs of images, mapping room programs in color to adequate
furniture layouts. The program retains wall structure and fenestration during image
translation while filling the rooms with relevant furniture, specified by each room’s
program. Figure 6 displays some typical results.
4 UI and Experience
We provide the user with a simple interface for each step throughout our pipeline. On
the left, users can input a set of constraints and boundaries to generate the resulting
122
S. Chaillou
Fig. 6 Results of model III. Source Author
plan on the right. The designer can then iteratively modify the input on the left to
refine the result on the right. Figure 7 showcases this type of interface & process, set
up for Model II.
5 Model Chaining and Apartment Building Generation
We scale the utilization of GANs in this part to the entire apartment building design.
The project uses an algorithm to chain Models I, II, and III, one after the other,
processing multiple units as single images at each step. Figure 8 shows this pipeline.
The challenge of drawing floor plates hosting multiple units marks the difference
between single-family houses and apartment buildings. Strategically, the ability to
control the position of windows and units’ entrances is key to enable unit placement
while ensuring each apartment’s quality. Since Model II takes doors and windows
position as input, the generation stack described above can scale to entire floor plates
generation.
The user is invited to specify the unit split between Model I and Model II, in other
words, specifying how each floor plate divides into apartments and to position each
unit entrance door and windows, as well as potential vertical circulations (staircases,
cores, etc.). The proposed algorithm then feeds each resulting unit to Model II (results
shown in Fig. 9), and then III (result in Fig. 10), to finally reassemble each floor plate
of the initial building. The algorithm finally outputs as individual images, all floor
plates of the generated building.
ArchiGAN: Artificial Intelligence x Architecture
123
Fig. 7 Design iterations using model II interface. Source Author
Fig. 8 Apartment building generation pipeline. Source Author
6 Going Further
If generating standard apartments can be achieved using this technique, pushing the
boundaries of our models is the natural next step. GANs offer remarkable flexibility
to solve seemingly highly constrained problems. In the case of floor plan layout,
partitioning and furnishing the space by hand can be a challenging process as the
124
S. Chaillou
Fig. 9 Model II output, program for each floor plate. Source Author
Fig. 10 Model III output, furnishing of each individual unit. Source Author
footprint changes in dimension and shape. Our models prove to be quite “smart” in
their ability to adapt to changing constraints, as evidenced in Fig. 11.
The ability to control the units’ entrance door and windows position, coupled
with the flexibility of my models, allows us to tackle space planning at a larger scale,
beyond the logic of a single unit. In Fig. 12, we scale our pipeline to entire buildings
generation while investigating our model’s reaction to odd apartment shapes and
contextual constraints.
7 Limitations and Future Improvements
If the above results lay down the premise of GANs’ potential for Architecture, some
clear limitations will drive further investigations in the future.
First, as apartment units stack up in a multi-story building, we cannot guarantee for now the continuity of load-bearing walls from one floor to the next.
Since all the internal structure is laid out differently for each unit, load-bearing walls
ArchiGAN: Artificial Intelligence x Architecture
125
Fig. 11 Flexible building layouts. Source Author
might not be aligned. For now, we consider the façades to be load bearing. However,
the ability to specify load-bearing elements’ position in the input of Model II could
potentially help address this issue.
Additionally, increasing the size of the output layer by obtaining larger
images which offer better definition is a natural next step. We want to deploy the
Pix2Pix HD project developed by NVIDIA in August 2018 to achieve this.
Finally, a major challenge comes from the data format of our outputs. GANs
like Pix2Pix handle only pixel information. The resulting images produced in our
pipeline cannot, for now, be used directly by architects & designers. Transforming
this output from a raster image to a vector format is a crucial step for allowing
the above pipeline to integrate with common tools & practices.
8 Conclusion
AI will soon massively empower architects in their day-to-day practice. As such
potential is about to be demonstrated, our work participates to the proof of concept
while our framework offers a springboard for discussion, inviting architects to
start engaging with AI, and data scientists to consider Architecture as a field of
investigation. However, today, our manifesto could be summarized in four major
points.
Conceptually first, our belief is that a statistical approach to design conception
shapes AI’s potential for Architecture. Its less-deterministic and more-holistic character is undoubtedly a chance for our field. Rather than using machines to optimize a
set of variables, relying on them to extract significant qualities and mimicking them
all along the design process is a paradigm shift.
Second, we are directionally convinced that our ability to design the right pipeline
will condition AI’s success as a new architectural toolset. Our choice for the “Grey
126
S. Chaillou
Fig. 12 GAN-enabled building layouts. Source Author
Boxing” approach, as introduced by Prof. Andrew Witt in Log, will likely secure the
best potential results. This method contrasts with the “black box” approach, that only
allows users to input information upfront, and to get finished design options at the
end of the process, without any control over the successive generation steps. To the
contrary, by breaking out our pipeline into discrete steps, “Grey Boxing” permits the
user to intervene all along the way. His tight control over the machine is his ultimate
guarantee of the design process quality.
Third, we technically believe that the sequential nature of the application will facilitate its manageability and foster its development. The ability to intervene throughout
ArchiGAN: Artificial Intelligence x Architecture
127
the generating process is a fundamental dimension: as each step of the pipeline represents a distinct portion of architectural expertise, each model can be trained independently, opening the way to significant improvements and experimentation in the
near future. Indeed, improving this entire pipeline end-to-end could be a long and
cumbersome task, while amending it step by step remains a manageable process,
within the reach of most architects and engineers in the industry.
Finally, we hope our framework will help address the endless breadth and
complexity of the models to be trained and those used in any generation pipeline.
Tackling parcels-footprint-room split-etc., as we do is one possible approach among,
we believe, a large set of options. To encapsulate the necessary steps of space planning, the key is more the principle than the method. And with the growing availability
of architectural data, we encourage further work and open-minded experimentation.
Far from thinking about AI as the new dogma in Architecture, we conceive this
field as a new challenge, full of potential, and promises. We see here the possibility
for rich results, that will complement our practice and address some blind spots of
our discipline.
References
1. Zheng, H., & Huang, W. (2018). Architectural drawings recognition and generation through
machine learning. Cambridge: MA, ACADIA.
2. Peters, N. (2017). Master thesis: “Enabling alternative architectures: Collaborative frameworks
for participatory design.” Cambridge, MA: Harvard Graduate School of Design.
3. Martinez, N. (2016). Suggestive drawing among human and artificial intelligences. Cambridge,
MA: Harvard Graduate School of Design.